In November's episode of Show & Tell, we spoke to Lam Thuy Vo and Sam Rabiyah of Documented, a nonprofit newsroom covering New York City's immigrant communities. Vo and Rabiyah walked us through their two-part series Decoding Immigrant TikTok and the associated pipeline they published on GitHub.
Vo and Rabiyah started by building what Rabiyah calls “the immigrant TikTok leaderboard,” a sample of roughly 10,000 of the most-watched TikTok videos about immigration in New York over 20 months. To do this, they relied on data from NYU's Center for Social Media, AI, and Politics — but their technical guide also provides advice on how to collect TikTok videos through data donations.
Their first pass surfaced that the most-viewed videos were from immigration lawyers and news influencers. The immigration lawyer videos came from a mix of real attorneys and copycat accounts that reuploaded their content, some of them using AI deepfakes. Scammers ran several of these accounts and collected fake legal fees through Cash App and Zelle. Some of the copycats in Documented’s sample outperformed the lawyers they were impersonating, according to Rabiyah.
The second story looked at Spanish-language “newsfluencers” who use selective framing and repetition to turn individual crimes into a narrative that NYC is a dangerous city overrun by immigrant criminals.
You can watch the episode above or on our YouTube channel.
In the “Show” section of the episode, Rabiyah walked us through the pipeline Documented built to analyze its trove of videos. The code is public on GitHub.
The pipeline includes a scraper that starts from a list of TikTok URLs, transcription via Whisper, translation with Google Translate, and a clustering step that groups videos by “containment” and “semantic” similarity. That last step lets you find videos that closely resemble each other, which can flag reuploads, edits, or videos pushing the same narratives. Every part of the pipeline is free and can run on your device.
Vo also shared practical advice on how to ethically collect and process TikTok data donated by readers. She said ongoing consent is key. So are a data retention policy and prewritten scripts, tested on your own data first, that minimize how much you handle the raw files.
About Show & Tell
Show & Tell is a monthly podcast hosted by Indicator co-founders Alexios Mantzarlis and Craig Silverman.
We talk about digital investigations with leading journalists, OSINT investigators, threat analysts, and trust and safety researchers. We aim to be as practical as possible by having our guests walk us through one or more investigations or analyses they actually worked on.
This is our tenth episode. Our previous episodes featured:
Shayan Sardarizadeh, senior journalist at BBC Verify.
Renée DiResta, associate research professor at Georgetown.
Caitlin Ostroff and Neil Mehta, investigative reporters at The Wall Street Journal.
Hal Triedman, a privacy and security researcher at Cornell Tech.
Omer Benjakob, disinformation and cyber reporter for Haaretz.
Barbara Marcolini, visual investigator with Amnesty International's Evidence Lab.
Kolina Koltai, senior researcher and trainer at Bellingcat.
Clara Jiménez Cruz, CEO of Maldita.es.
Jeff Horwitz, investigative reporter for Reuters.
We hope you enjoy and learn from the conversation with Lam and Sam. Catch every episode by subscribing to our YouTube channel. We're also on Spotify and Apple Podcasts.
Have a suggestion for who we should feature next? Email us.
